Overview

Goal is to assess the noise measurements from passby measurements in 2018, where four quiet pavement treatments were applied to test areas: 1/4 in chip seal, 3/8 in chip seal, Type II microsurfacing, and Type III microsurfacing.

There are eight sites, and four pavement treatments; each pavement treatment is replicated at two sites. Sound intensity was measured pre-treatment and post-treatment.

In the original assessment, overall the Type II microsurfacing was found to reduce overall sound intensity sound from OBSI measurements from 99.2 dB on average to 97.6. Other treatments did not on average reduce sound levels.

This analysis focuses on the observational passby data, where vehicle type varies, and assesses the effects of speed and pavement temperature in addition to experimental treatments.

In the table below, SD_pavetemp is not applicable when the value is 0, because the same pavement temperature measurement was used throughout the trial. 2019 data are shown but only 2016 and 2018 data are used for statistical analyses. These analyses exclude non-passenger car vehicle types (i.e., bus or motorcycle measurements), which occurred infrequently and therefore vehicle type could not be used as a covariate in the analyses.

Summary of passby data collected by year and treatment.
Year Treatment N total N autos Mean speed SD speed Mean pavement temp SD pavement temp
2016 Baseline - Chip Seal 95 93 53.47 7.13 95.59 4.99
2018 Type II Microsurfacing 28 27 45.36 5.55 124.16 1.27
2018 Type III Microsurfacing 17 16 51.24 9.61 69.26 3.84
2018 1/4 in Chip Seal 15 15 54.33 8.04 89.80 3.90
2018 3/8 in Chip Seal 30 30 45.53 6.58 119.00 0.00
2019 Type II Microsurfacing 34 34 45.09 6.34 66.76 16.02
2019 Type III Microsurfacing 24 14 56.04 5.61 81.00 0.00
2019 1/4 in Chip Seal 37 37 48.78 6.76 64.00 0.00
2019 3/8 in Chip Seal 34 34 43.76 6.08 85.00 0.00

2018 MANOVA Analysis with LZFeq

Using MANOVA, we see that there are strong effects of all the predictors on noise values across the frequencies. The table shows the summary of the statistical test for the difference in sound intensity across frequencies attributable to each of the predictors: pavement temperature, speed, treatment, and the statistical interactions between temperature and speed, and between treatment and speed. All predictors were statistically significant, including the interactive effects of speed and temperature, and speed and treatment.

Pillai’s trace is a test statistic used in multivariate analyses. It is based on the eigenvalues associated with each predictor, across all the sound intensity frequencies. A larger value indicates that this predictor explains more of the difference in the response data (the matrix of all the sound intensity levels across frequencies). In order to interpret the statistical significance of this test statistic, we examine the p-value. This represents whether the test statistic is larger than expected by chance; a p-value less than 0.05 is considered statistically significant.

Frequencies from 50 Hz to 20 kHz were used for this analysis, covering 27 1/3 octave band frequencies.

Summary of MANOVA analysis for 2018 treatments compared to 2016.
Df Pillai approx F num Df den Df Pr(>F)
treatment 4 2.849 13.748 108 600 < 0.001
pavetemp 1 0.287 2.197 27 147 0.002
speed 1 0.683 11.755 27 147 < 0.001
pavetemp:speed 1 0.245 1.762 27 147 0.018
treatment:speed 4 1.058 1.998 108 600 < 0.001
Residuals 173

Post-hoc ANOVAs by LZFeq Frequency

Plotting ANOVA change in sound intensity from pre-treatment for LZFeq.

Response variables: treatment, pavement temperature, and speed, as well as the interactions between treatment and speed, pavement temperature and speed.

The plots are generated from the output of the statistical models, holding pavement temperature constant at 90 degrees and speed constant at 50 mph. The change in sound intensity is calculated as the difference between the predicted sound intensity at 90 degrees and 50 mph for a particular treatment, compared to the baseline pavement conditions in 2016 in the same conditions. The error bars are the standard errors for predicted sound intensity levels.

The predicted values at these conditions are saved as LZFeq_ANOVA_PredictedVals.csv and LZFeq_ANOVA_PredictedVals_StdErr.csv

Compared with Figure 18 in DEVA_QPP_Report_Nov2106-May-2018_draft2 9-5-8.pdf.

Plotting standardized curves

The following shows the predicted sound intensity levels at standardized pavement temperature (90 degrees) and vehicle speed (50 mph). Two versions are shown, a point plot and a ‘ribbon plot’. Both show the predicted values, +/- 1 standard error.

These plot are interactive (static versions can be made as well); clicking on the legend will show or hide a specific treatment.

2019 MANOVA Analysis with LZFeq

For the 2016 to 2016 comparison, all predictors remained significant, except that the interaction between pavement temperature and speed became only marginally significant (p = 0.076). The variable is retained here for comparison with the previous analysis of 2016 to 2018 data.

Summary of MANOVA analysis for 2019 treatments compared to 2016.
Df Pillai approx F num Df den Df Pr(>F)
treatment 4 2.383 10.315 108 756 < 0.001
pavetemp 1 0.223 1.982 27 186 0.004
speed 1 0.685 14.946 27 186 < 0.001
pavetemp:speed 1 0.175 1.461 27 186 0.076
treatment:speed 4 0.707 1.502 108 756 0.001
Residuals 212

Post-hoc ANOVAs by LZFeq Frequency

Plotting ANOVA change in sound intensity from pre-treatment for LZFeq, comparing 2019 measurements to 2016.

Plotting standardized curves

The following shows the predicted sound intensity levels at standardized pavement temperature (90 degrees) and vehicle speed (50 mph). Two versions are shown, a point plot and a ‘ribbon plot’. Both show the predicted values, +/- 1 standard error.

These plot are interactive (static versions can be made as well); clicking on the legend will show or hide a specific treatment.